Research On Highway Guardrail Detection Method Based On Video

With the development of transportation,traffic congestion and safety become increasingly serious.Intelligent transportation system(ITS)has become one of the important means to solve these problems.In the research of ITS,the perception of traffic environment around vehicles is the basis of vehicle decision-making and control.Highway guardrail detection system is an indispensable part of intelligent vehicle environment perception,and plays an important role in ITS.In the research of guardrail detection algorithms based on images,the groove feature of highway guardrail was ignored,and the single feature detection method was used mostly.Based on the current highway guardrail detection algorithms,and considering the real-time of the algorithm and the practicability of intelligent vehicles,two detection algorithms are proposed in this thesis: highway guardrail detection based on steerable filters,and highway guardrail detection based on features fusion.The main work accomplished in this thesis as follows:1.Analyzing the situations of highway guardrail detection,the deficiencies of conventional algorithms are summarized,and the topic of this thesis is put forward.Then,the related technologies and algorithms used in this thesis are summarized.2.A detection algorithm of highway guardrail based on steerable filters is studied.Firstly,the template of the guardrail is determined by applying the steerable filters,and using the adaptive template method,the matching detection of the guardrail can be achieved.Then,the model of highway guardrail is constructed,the parameters of model are obtained by fitting the matching results.And the detection algorithm of the highway guardrail based on the steerable filters is established.3.A detection algorithm of highway guardrail based on the fusion of local binary pattern(LBP)feature and histogram of oriented gradient(HOG)feature is studied.Firstly,the features of LBP and HOG of the guardrail in the image are extracted.Secondly,the LBP and HOG features are fused and the dimensionality is reduced,then the guardrail is detected by support vector machine(SVM)algorithm.Finally,the detection accuracy is improved by using different sizes of sliding windows and clustering fitting.4.Using the data sets of KITTI,POSS and Google,and comparing with several detection methods,the effectiveness of the detection methods proposed in this thesis is verifiedThe experimental results show that the detection methods proposed in this thesis have some improvements in detection performance compared with traditional detection methods.Especially,the improvement of the detection algorithm based on features fusion is more obvious than the detection method based on single feature.The research work presented in this thesis has a certain reference value for the future research of highway guardrail detection.